798 Journal of Power Electronics, Vol. 16, No. 2, pp.

798-804, March 2016

http://dx.doi.org/10.6113/JPE.2016.16.2.798
JPE 16-2-38 ISSN(Print): 1598-2092 / ISSN(Online): 2093-4718

A Control Method for Power-Assist Devices using a

BLDC Motor for Manual Wheelchairs
Dong-Youn Kim*, Yong-Hyu Kim**, Kwang-Sik Kim***, and Jang-Mok Kim†
*,†
Department of Electrical Engineering, Pusan National University, Busan, Korea
**
Busan Techno Park, Busan, Korea
***
LG Electronics, Changwon, Korea

Abstract
This paper proposes a new operation and control strategy for Power-Assisted Wheelchairs (PAW) using one brushless DC
(BLDC) motor. The conventional electrical wheelchairs are too heavy and large for one person to move because they have two
electric motor wheels. On the other hand, the proposed PAW system has a small volume and is easy to move due to the presence of a
single wheel motor. Unlike the conventional electric wheelchairs, this structure for a PAW does not have a control joystick to reduce
its weight and volume. To control the wheelchair without a joystick, a special control system and algorithm are needed for proper
operation of the wheelchair. In the proposed PAW system uses only one sensor to detect the acceleration and direction of PAW’s
movement. By using this sensor, speed control can be achieved. With a speed control system, there are three kinds of operations that
can be done on the speed of a PAW: the increment of PAW speed by summing external force, the decrement of PAW speed by
subtracting external force, and emergency breaking by evaluating the time duration of external force. The validity of the proposed
algorithm is verified through experimental results.

Key words: Assist equipment, Brushless DC motor, Power-assisted wheelchair sensor, Speed control, Wheelchair control

of PAW systems. One is the systems having two motors

I. INTRODUCTION
In recent years, elderly and disabled people have become
more active in social activities due in part to the development
of medical and electric technology. The areas of activity for
these people are wider than they were a decade ago. Electric
wheelchairs are usually used for short distance movement
while cars are required for the movement of electric
wheelchairs over a long distance. However, conventional
wheelchairs which have joysticks are too heavy and large to
load into a car alone. Some PAWs have been developed to
achieve light weight and portability from manual wheelchair
[1]-[9]. These PAWs are not equiped with joysticks and use
torque sensors or acceleration sensors to measure or estimate
the speed or acceleration which is generated by the pushing
power of the user. Pushing power is used for the propulsion
of manual wheelchairs. For the most part, there are two types
Manuscript received Apr. 6, 2015; accepted Oct. 5, 2015
Recommended for publication by Associate Editor Kwang-Woon Lee.
†
Corresponding Author: jmok@pusan.ac.kr
Tel: +82-51-510-2366, Fax: +82-51-513-0212, Pusan Nat’l University
* and compared to the conventional PAW control algorithm.
Dept. of Electrical Eng., Pusan National University, Korea
**
Busan Techno Park, Korea
***
LG Electronics, Korea
© 2016 KIPE
connected with both wheels of the wheelchair. In this type of
PAW, since the PAW is already installed on the wheelchair,
there is no need for the user to install it. However, it is
difficult to repair once it is damaged. The second type is the
attachable power assist device (PAD), where attachable and
portable hardware is attached to a wheelchair whenever the
user wants to use the PWA system. Attachable PADs can be
simply applied to several manual wheelchairs.
The conventional studies on PAWs focused on the use of
two motors. These PAW systems use pushing power to
generate the assist power. Because assist power is
temporarily added for propulsion, a number of propulsion
motions (pushing wheels) are required to continuously move
the wheelchair. However, excessive propulsion motions have
a bad effect on wrist joints [10]. In addition, the installation
of two motors causes a weight increment. Therefore, the user
needs considerable force to load or unload a wheelchair into
or out of a car.
In this paper, a different control algorithm is proposed
The concept of the proposed algorithm is that the wheelchair
 A Control Method for Power-Assist … 799

y v y v
Өl
v v
ω

Өr R
W ω
R Өr ω
Φ
x x
Fig. 1. Electrical wheelchair model. Fig. 2. Proposed PAW wheelchair model.

can maintain speed for a long time with a single pushing

action. It helps the user move a long way without applying a Tr  J  J
R  
W
l  r  (6)
lot of pushing actions. This proposed control algorithm uses
To derive the motion equation of the wheelchair, the total
one acceleration sensor. The control algorithm is composed
kinetic energy of the wheelchair L is given by equation (7).
of three control methods. One is the accumulation of the
1 mR 2  1 R 1
speed command to increase the speed of the wheelchair. The L ( l  r ) 2  J ( (l  r )) 2  J  (l 2  r 2 )
second method is the deceleration of the speed command to 2 4 2 W 2
(7)
decrease speed of the wheelchair. The third is the quick stop
Solving the Lagrange equation, the load torque equation of
control for emergency situations. These algorithms are based
the wheelchair can be presented as (8).
on the output of the acceleration sensor and are combined for
speed control. The proposed algorithm is verified by  mR 2 JR 2 mR 2 JR 2 
 2  J  2    
 r   4  
experimental results.
    W 4 W   r   Bw  r 
  mR 2
JR 2 mR 2
JR 2
   l 
 l
  2  2  J    l 
 4 W 4 W 
II. MODELING OF THE PROPOSED PAW (8)
WHEELCHAIR
where, m and J are the mass and inertia of the wheelchair,
A. Conventional Model of a Wheelchair Controlled by Two and J  and B are the inertia and friction coefficient of
Wheels [11] the wheel, respectively.
Fig. 1 shows the conventional model of a wheelchair. The
wheel chair model has two domains which are the propulsion B. Model of the Proposed PAW Wheelchair
and rotation, where v is the propulsion speed, and  is the Fig. 2. shows the proposed model using one propulsion
rotational speed. The kinematics of the wheelchair can be motor. In this case, the motor can control only the propulsion
represented as (1) and (2). force. That means the rotational force can be neglected, that
is l and  r are equal. The kinematics of the proposed
R

v  l  r
2
 (1) wheelchair can be represented as (9) and (10).
v  R 

W
R 

 l  r  (2)
v  R 
(9)
(10)
where l , r are the angular speeds of the left and right where,  and  are the angular speed and acceleration
wheels,  is the moving direction, R is the radius of the speed of the propulsion motor.
wheel, and W is the distance between the two wheels. The propulsion force is replaced by (11), and the total kinetic
Then, the time derivative of (1) and (2) can be represented as energy of the proposed wheelchair L is given by equation
follows. (12).
R
2

v  l  r  (3) F p  mv  mR (11)

 
W

R  
l   r  (4) L
mR 2  2 J   2
2
 
2
 (12)
The propulsion force and rotation torque of the wheelchair Finally, the load torque equation of the proposed wheelchair
can be expressed as (5) and (6). can be derived as (13).
R
2

F p  mv  m l  r  (5)   mR2  J   B 
  (13)
800 Journal of Power Electronics, Vol. 16, No. 2, March 2016

III. PROPOSED POWER ASSIST DEVICE CONTROL

ALGORITHM vh
A. Maximum Speed Tracking and Keeping Algorithm
In the proposed PAW, an acceleration sensor is used to
ah
detect the external force applied by the user. The output of
the acceleration sensor is converted to the additional
speed, v h , which is used for the generation of the speed
command of the proposed PAW system. v
The acceleration sensor has a positive value when the user
Fig. 3. Movement of Power-Assisted Wheelchair.
pushes the wheelchair wheels in the positive direction as
shown in Fig. 3. Fig. 3 shows the direction of the additional Maximum
speed and acceleration when a positive direction external point
v va_max
force is applied, where a h is the acceleration measured by va
the acceleration sensor, and v h is the additional speed vh

produced by an external force. v h can be calculated by

integrating the acceleration as given by equation (14).
t
vh   ah dt (14) ah
0
0 t
This additional speed can be added to or subtracted from
the real wheelchair speed, v . Fig. 4. Profile of acceleration sensor and detecting maximum
Fig. 4 shows the profile of ah and vh. When the user speed to keep constant speed for PAW.
pushes the wheelchair wheels, ah can be measured through
the acceleration sensor. Following the profile of ah, the real
velocity of the wheelchair increases. However, the speed of
the wheelchair steadily decrease due to frictional force. The
external force makes the additional speed profile vh like the
half cycle of a sine waveform. From the profile,
assisted-speed va can be generated from vh by using equation
(15). It is not suitable to adapt vh as a speed command from
the acceleration sensor output since it has a lot of noise and a
small magnitude. Thus, the scale and slope of va can be
controlled by using equation (15).
Ka Fig. 5. Summation of speed by adding the external force.
va  vh (15)
1  a s
where, va is the assisted-speed, K a is the assistance where v a _ old is the value of v a calculated before a period
coefficient, and  a is the time constant. of time.
To drive continuously by applying only one propulsion B. Algorithm for Increasing the Speed Command
motion, the maximum value is retained for the calculation of
In the proposed PAW system, the speed of the wheelchair
the speed command. The speed command va* can be can be changed according to the magnitude and direction of
acquired by using the differential value of va to detect the the external force. Therefore, the speed command can be
maximum speed value. Fig. 4 schematically shows the above determined by summing the number of the external forces
mentioned explanations of the mechanism used to detect the according to the drive direction as va1 , va2 , va3 and
maximum speed point. According to the differential value of va4 which are determined by each external force as shown in
va , the speed command va* can be presented like equation Fig. 5. This means that whenever an external positive force is
(16) applied, the speed command increases to the next level as
 va va  va _ old  0 depicted in Fig. 5.
va *   (16)
 a_max  va _ old
v va  va _ old  0 C. Braking Algorithm
 A Control Method for Power-Assist … 801

A braking algorithm must be included in the proposed

system unlike conventional electric wheelchairs with a
joystick or PAW based torque controls. vb
There are two steps in the proposed braking algorithm.
ah
Firstly, the brake ratio is decided according to the time
duration. Then, the braking method is decided since there are
two braking methods. One is the deceleration in case of a
decrement in the speed command following the decided brake v
ratio. Another is a quick stop if the user wants to reduce the Fig. 6. The velocity and acceleration during deceleration/stop.
speed command quickly.
The braking motion is activated by grasping the wheelchair
wheel. In this case, the profile time of the deceleration is the
reference signal of the braking as shown in Fig. 6 and Fig. 7,
respectively.
Fig. 6 shows the direction of the additional speed and the
acceleration of the wheelchair during the braking operation.
The acceleration has a negative value in accordance with the
driving direction.
Fig. 7 shows the acceleration pattern in case of the braking
operation of the wheelchair. To avoid the effects of noise, the
brake signal is recognized after a few seconds from the point Fig. 7. Acceleration during reduction of speed.
where grasping of the wheel occurs.
In Fig. 7, t1 is the point where grasping the wheelchair
0 < Tbr < α Tbr > α
wheels for braking occurs. It is decided that the acceleration
Brake ratio = Brake_ratio =
signal goes down under zero. t2 is the point of distinction large value small value

for brake mode: speed decrement or complete break. The Brake_point t Brake_point t
distinction is made up with the speed error and acceleration (a) (b)
signal. If the user grips the rim strongly, the speed error Fig. 8. Speed command according to the brake ratio. (a) Large
reaches the distinction point in a short time. Otherwise, if the brake ratio. (b) Small brake ratio.
user grips the rim smoothly and continuously, the speed error
gently increases to the distinction point over a long time. The
distinction time is experimentally defined to be 15% of the
speed commands. Thus, Tbr is the detection time between
t1 and t2 and it is related to the brake ratio. The brake ratio
determines the decremental ratio of the speed command. If
Tbr is short, it means that the user wants to quickly reduce the
speed command.
Fig. 8 shows the speed command according to Tbr and
the brake ratio. The variables α is used for the division of (a) Speed command with the acceleration and deceleration.
Tbr . If Tbr has an arbitrary value between 0 and α, as in Fig.
8(a), it can be considered that the user wants to stop quickly.
In this case, the brake ratio is determined to be a large value.
If Tbr is longer than α, as shown in Fig. 8(b), the brake ratio
is determined to have a small value and α is set by the
acceleration profile.
After the brake ratio is determined, Tstop is used for
selecting the deceleration or quick stop of the wheelchair.
Tstop is the elapsed time between t2 and t3 . If the user
(b) Speed command with the acceleration and quick stop.
grasps the wheel for a moment, the speed is reduced for a
short time and the speed again follows the speed command. Fig. 9. Speed command about two cases of brake.
802 Journal of Power Electronics, Vol. 16, No. 2, March 2016

DSP Control System

< Speed Command System >

< Assist speed generation >

Ka dva
1  τas dt
i*
Speed Generation Speed Accumulation
v Tracking Keeping
v va*
va*
t t va* i < PWM Inverter > <Propulsion Action>
v* `
< Brake System > Variable brake ratio
v 0 <Tbr<  Kbvb ia sensor
 <Tbr Kb . M
ib
Tbr
t1 t2
Tstop
v= A
X ic
α β t3 t(s) t

Stop / Deceleration
.
ah v
a(m/s2) - vb
0 < Tstop <
- vb < BLDC motor >

Brake Point Detection  <Tstop

t

Brake Operation < Acceleration sensor >

PAW hardware – wheelchair & inverter

Fig. 10. Whole block diagram of the proposed control algorithm.

In this case, the acceleration acquires negative value for a

moment. However, if the user grasps the wheel for a long
time, the speed of the wheelchair is reduced continuously. In
this case, the acceleration has negative value for a long time.
If Tstop is longer than β, the system will decide to quick stop
the wheelchair. Otherwise, it can be considered to be a
deceleration of the speed command.
The deceleration operation is implemented in stages like
(a) Configuration of PAW (b) Backside of Wheelchair
the increasing operation of the speed command. As shown in
Fig 9(a), the speed command is reduced in stages during the Fig. 11. Configuration of wheelchair.
deceleration operation. The quick stop operation is used for
quick stops in case of an emergency. Once the brake signal
occurs, the first stage of deceleration is implemented. As
shown in Fig. 9(b), the quick stop signal is generated during
the first deceleration step.

IV. EXPERIMENTAL RESULTS (a) (b)

Fig. 12. DSP control board and acceleration sensor. (a) DSP
Fig. 10 shows a whole block diagram of the proposed control board used for the experiment. (b) Acceleration sensor
algorithm. The proposed PAW system is composed of a DSP AM-3AXIS V03.
control system, a manual wheelchair and a PAW hardware
system. The acceleration generated by an external force is Fig. 11(a) shows the configuration of the wheelchair on
used as the speed command for the DSP control system, and which the power assist device is installed. Fig. 11(b) shows
the system keeps the speed command steadily. If additional the backside of the wheelchair with the PAD. The power
acceleration occurs during the speed control, the speed rating of the motor is 300(W).
command increases in step. When the braking motion is Fig. 12(a) shows a control board which uses a
detected, the brake-command ( K b vb ) is subtracted from the TMS320F335 as the main processor [12]. The acceleration
sensor for the proposed algorithm is represented in Fig. 12(b).
speed command ( v a* ). Then, the final speed-command ( v * ) The sensor is a AM-3AXIS of STMicroelectronics [13] and
is used as the reference speed for the DSP control system. experimental data are logged by a NI DAQPAD-6251.
 A Control Method for Power-Assist … 803

1.2m/s v*(speed command) 1.6m/s

v* (89rpm) Brake ratio 1.2m/s
(speed cpmmand) = 0.01
v(real speed)
0.6m/s
t3_cnt 2000 0
0
10A
i(current)
ah 5A
(acceleration)
-1m/s2
0.5s 0
Fig. 13. Quick stop operation when Brake ratio = 0.01.
0
1.2m/s ah threshold -0.3m/s2
v*(89rpm) (acceleration) -0.6m/s2
(speed cpmmand) Brake ratio
= 0.005
5s
7800
Fig. 16. Experimental results of the continuous decrement of the
t2_cnt
speed.
0

ah v* 1.6m/s
(acceleration)
-0.75m/s2 (speed command)
1.2m/s
0.5s

Fig. 14. Quick stop operation when Brake ratio = 0.005. 0.6m/s
v(real speed)
0
1.6m/s 0.6m/s2
1.2m/s
v*(speed command) v 0
(real speed) 0.6m/s i(current)
-0.6m/s2
0
10A 2s
i(current) Stop
Count 20000
5A
0

0
5s
0.6m/s2
Fig. 17. Experimental result of the quick stop of the speed.

threshold
ah(acceleration)
5s
Fig. 15. Experimental results of the continuous increment of the
speed.
Fig. 13 and Fig. 14 show experimental results of the brake
ratio according to Tbr . In the experiment, α is 0.5s. The Tbr
counter counts every 100us. As shown in Fig. 13, the brake
ratio is determined to have a large value when Tbr is shorter
than 0.5s. As shown in Fig. 14, the brake ratio has a small
value because Tbr is longer than 0.5s.
0
command ( v * ), the current increases during the transient
state, and the current is maintained at a fixed speed.
Fig. 16 shows a waveform during the continuous
deceleration operation of the wheelchair. To avoid
malfunctions, the brake point is recognized when the
acceleration is lower than the threshold. The current (i)
increases during the deceleration operation. The reason is that
a motion which grasps the wheelchair wheel is considered as
an increase of the load condition until the controller
recognizes the brake point.
Fig. 17 shows a waveform during quick the stop operation.
In the experiment, the signal of the quick stop is generated
when the acceleration ( a h ) has negative value for more than

Fig. 15 shows experimental result of the continuous
propulsion operation. To avoid malfunctions caused by noise,
the propulsion motion is recognized when the output of the
sensor is larger than a threshold value. In this experiment, the
maximum speed is limited to 1.6m/s for the sake of safety,
and the speed of each stage is 0.6 m/s. To track the speed
two seconds (β=2s) from the starting point of the first
deceleration step. In Fig. 15, the stop count increases every
100us. If the stop count equals to 20000, the speed command
is reduced to zero directly. The real speed ( v ) cannot follow
the speed command ( v * ) because the user grasps the
wheelchair wheels to reduce the speed.
804 Journal of Power Electronics, Vol. 16, No. 2, March 2016
V. CONCLUSION
In this paper, new operation and control strategies for the
PAW of a manual wheelchair was proposed. The motion
recognition was implemented in the PAW control system by
the detection of the acceleration sensor output signal. The
speed of the wheelchair was increased step by step according
to external force applied by the user. In the brake operation,
the user reduces the speed step by step or quickly and the
controller determines whether the step by step or quick speed
reduction is required according to the time duration of the
acceleration. The validity of the proposed algorithm was
verified through experimental results.
APPENDIX
Motor Data Wheelchair Data
Stator Resistor 0.25 [Ώ] Wheelchair Dimension
Stator Inductance 565[uH] (L x W x H)
Rated output power 300[W] Wheelchair weight
Rated current 9[A] Wheel radius
Rated voltage 36[V] Power Assist Device Dimension
Number of poles 60 Power Assist Device Weight
ACKNOWLEDGMENT
This research was financially supported
INNOPOLIS Foundation (14BSI388)
REFERENCES
[1] H. Seki, T. Sugimoto, and S. Tadakuma, “Novel straight
road driving control of power assisted wheelchair based on
disturbance estimation and minimum jerk control,” in
Industry Applications Conference, Vol. 3, pp. 1711-1717,
Oct. 2005.
[2] P.-W. Hsueh, M.-C. Tsai, H.-T. Pan, and A. Grandjean
“Integrated synchronized motion control for a force
sensorless power-assisted wheelchair,” in Preprints of the
18th IFAC World Congress, Vol. 18, pp. 5956-5961,
Aug./Sep. 2011.
[3] Y. Munakata, A. Tanaka, and M. Wada, “An active-caster
drive system for motorizing a manual wheelchair,” in
IEEE International Conference on Mechatronics and
Automation(ICMA), pp. 1161-1166, Aug. 2013.
[4] D. K. Jones, R. A. Cooper, S. Albright, and M. Digiovine,
“Power wheelchair driving performance using force and
position sensing joysticks,” in Proc. IEEE Bioengineering
Conference, pp. 130-132, Apr. 1998.
[5] Y. Oonishi, S. Oh, and Y. Hori, “A new control method
for power-assisted wheelchair based on the surface
myoelectric signal,” IEEE Trans. Ind. Electron., Vol. 57,
No. 9, pp. 3191-3196, Sep. 2010.
[6] C. Ou, C. Chen, and T. Chen, “Modeling and Design a
Power Assisted Wheelchair used Torque Observer,” in
International Symposium on Computer Communication
Control and Automation, Vol. 2, pp. 63-66, May 2010.
[7] Y. Munnakata, A. Tanaka, and M. Wada, “A five-wheel
wheelchair with an active-caster drive system,” in IEEE
International Conference on Rehabilitation Robotics
(ICORR), pp. 1-6, Jun. 2013.
[8] S.-W. Hwang, C.-H. Lee, and Y.-B. Bang, “Power-assisted
wheelchair with gravity compensation,” in 12th
International Conference on Control, Automation and
Systems(ICCAS), pp. 1874-1877, Oct. 2012.
[9] H. Seki and N. Tanohata, “Fuzzy control for electric
power-assisted wheelchair driving on disturbance roads,”
IEEE Trans. Syst., Man, Cybern., Part C: Applications
and Reviews, Vol. 42, No. 6, pp. 1624-1632, Nov. 2012.
[10] I. Uriostegui, F. Bernal, J. A. Tover, E. Vela, M. Bourdon,
and A. I. Perez, “Biomechanical analysis of the propulsion
of the manual wheelchair in patients with spinal cord
injury,” in Pan American Health Care Exchanges
(PAHCE), pp. 1-5, Apr. 2014.
[11] J. Miyata, Y. Kaida, and T. Murakami,
“v–˙φ-coordinate-based power-assist control of electric
wheelchair for a caregiver,” IEEE Trans. Ind. Electron.,
Vol. 55, No. 6, pp. 2517-2524, Jun. 2008.
[12] TMS320F28335 data manual, Texas Instrument, 2007.
[13] LIS344ALH Datasheet, STMicroelectronics, 2008.
Dong-Youn Kim was born in Ulsan, Korea,
in 1983. He received his B.S. and M.S.
degrees in Electrical Engineering from
Pusan National University, Busan, Korea, in
2011 and 2013, respectively, where he is
590 x 950 x 850[mm]
presently working towards his Ph.D. degree.
15.5[kg] His current research interests include power
0.13[m] conversion, electric machine drives, and
430 x 330 x 180[mm]
electrical vehicle propulsion.
17.8[kg]
Yong-Hyu Kim received his B.S. from
Kunsan National University, Gunsan, Korea,
in 2004; and his M.S. degree from the radio
by the science engineering, Chungnam National
University, Daejeon, Korea in 2012. He is
working toward the Ph.D. degree in
Electrical Engineering from Pusan National
University, Busan, Korea. From 2005 to
2008, he was an associate research engineer at Korea
Automotive Technology Institue (KATECH). Since 2008, he has
been research engineer of automotive parts technology support
center at the Busan techno park. His current research includes
component of electrical vehicle.
Kwang-Sik Kim was born in Busan, Korea,
in 1988. He received his M.S. degree in
Electrical Engineering from Pusan National
University, Busan, Korea, in 2015. He is
currently working for LG Electronics,
Changwon, Korea. His current research
interests include electric machine drives in
home appliances, and electrical vehicle
propulsion.
Jang-Mok Kim received his B.S. from
Pusan National University (PNU), Busan,
Korea, in 1988; and his M.S. and Ph.D.
degrees from the Department of Electrical
Engineering, Seoul National University
(SNU), Seoul, Korea, in 1991 and 1996,
respectively. From 1997 to 2000, he was a
Senior Research Engineer with the Korea
Electrical Power Research Institute (KEPRI), Daejeon, Korea.
Since 2001, he has been with the School of Electrical
Engineering, PNU, where he is presently a Research Member in
the Research Institute of Computer Information and
Communication, a Faculty Member, and a head of the LG
Electronics Smart Control Center. As a Visiting Scholar, he
joined the Center for Advanced Power Systems (CAPS), Florida
State University, Tallahassee, Florida, in 2007. His current
research interests include the control of electric machines,
electric vehicle propulsion, and power quality.